Re: secure execution of drivers
On Tuesday 25 November 2008 22:29, "Aneurin Price" <aneurin.price@gmail.com>
wrote:
> Based on my experience, I would not personally recommend XFS to anyone who
> cannot guarantee that their system will absolutely never crash or suffer
> power failure. XFS's failure modes seem pretty disastrous. Then again I
> never had any problems when I was using reiserfs, so YMMV.
One issue with XFS is that data which is not written synchronously (IE opened
with O_DIRECT or O_SYNC) or explicitly synchronised with fsync() may stay in
the write-back cache for up to 30 seconds (compared with 5 seconds for
Ext2/3).
Now for correctly written applications that call fsync() when appropriate this
should not be a problem. But for buggy applications this can cause
significant problems.
One example where this caused a disaster was a program that used the create
and rename method of replacing a file. It would often replace the file more
than once in 30 seconds. This meant that on XFS the main file and the backup
could end up without any data blocks but on Ext3 as the period was >5 seconds
it mostly worked. Of course the program in question worked better on Ext3 as
well once I put a fsync() call in the source...
--
russell@coker.com.au
http://etbe.coker.com.au/ My Main Blog
http://doc.coker.com.au/ My Documents Blog
Reply to: